Video-MME — raw "VideoMME"; Kimi K2.5 card; no w/ or w/o subtitle marker, mode ambiguous — kept separate from labeled variants
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Kimi K3 | Moonshot | 90 | 1 | 2026-08-24 |
| 2 | GPT-5.6 Sol | OpenAI | 89.5 | 1 | 2026-08-24 |
| 3 | Gemini 3 Pro | 88.4 | 1 | 2026-06-15 | |
| 4 | Qwen3.7 Plus Preview | Alibaba | 88 | 1 | 2026-08-23 |
| 5 | MiMo V2.5 Pro | Xiaomi | 87.7 | 1 | 2026-08-23 |
| 6 | Kimi K2.5 | Moonshot | 87.4 | 2 | 2026-08-23 |
| 7 | GPT-5.2 | OpenAI | 86 | 1 | 2026-06-15 |
| 8 | Claude Opus 4.8 | Anthropic | 86 | 1 | 2026-08-24 |
| 9 | MiniMax M3 | MiniMax | 85.4 | 1 | 2026-08-23 |
| 10 | Gemini 2.5 Pro | 84.8 | 1 | 2026-08-23 | |
| 11 | Qwen3.6 Plus | Alibaba | 84.2 | 1 | 2026-08-23 |
| 12 | Qwen2.5 VL 72B | Alibaba | 79.1 | 1 | 2026-06-05 |
| 13 | Qwen3 VL 235B A22B Reasoning | Alibaba | 79 | 1 | 2026-06-15 |
| 14 | Gemini 1.5 Pro | 78.6 | 1 | 2026-08-23 | |
| 15 | Qwen2.5 VL 32B | Alibaba | 77.9 | 1 | 2026-06-05 |
| 16 | GPT-4o | OpenAI | 77.2 | 2 | 2026-06-05 |
| 17 | Qwen3 VL 30B A3B Instruct | Alibaba | 74.5 | 1 | 2026-08-23 |
| 18 | Qwen3 VL 30B A3B Reasoning | Alibaba | 73.3 | 1 | 2026-08-23 |
| 19 | Kimi VL A3B Instruct | Moonshot | 72.7 | 2 | 2026-06-12 |
| 20 | Kimi VL A3B Thinking | Moonshot | 71.9 | 2 | 2026-06-05 |
| 21 | Qwen3 VL Thinking (8B) | Alibaba | 71.8 | 1 | 2026-08-23 |
| 22 | Qwen3 VL 8B Instruct | Alibaba | 71.4 | 1 | 2026-08-23 |
| 23 | GPT-4o mini | OpenAI | 68.9 | 1 | 2026-06-05 |
| 24 | Gemini 1.5 Flash 8B | 66.2 | 1 | 2026-08-23 | |
| 25 | Mage-VL-4B | Microsoft | 64 | 1 | 2026-07-26 |
| 26 | Gemma 3 12B | 62.1 | 2 | 2026-06-05 | |
| 27 | Qwen3 VL 4B (Reasoning) | Alibaba | 59.7 | 1 | 2026-07-26 |
| 28 | Phi 4 R V 15B | Microsoft | 55.3 | 1 | 2026-07-26 |
| 29 | Phi 4 Multimodal Instruct | Microsoft | 55 | 1 | 2026-08-23 |
| 30 | Phi 4 MM 5.6B | Microsoft | 44.7 | 1 | 2026-07-26 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.